Support for NVIDIA Tesla C1060 Computing Processor

Question

Level 1

4 points

Support for NVIDIA Tesla C1060 Computing Processor

I recently discovered that XFDTD, an electromagnetics simulation program, can utilize up to four NVIDA CUDA-enabled graphics cards to reduce computation time by orders of magnitude. While I have a single NVIDIA GeForce GTX 285, I'd be interested in adding one or more TESLA cards to my early 2009 Mac Pro.

Does anyone have any information on whether or not such a card will be supported in the future? Given that Apple computers are promoted as scientific computing platforms (by Apple and myself), I think supporting such a card is must. I also understand that the manufacturers must provide support for EFI and Apple must write the drivers.

The biggest advantage the Tesla card provides over the NVIDIA GeForce GTX 285 and Quadro FX 4800 cards is that it has 4 GB of GDDR3 RAM, instead of the 1 GB and 1.5 GB of the other cards respectively.

http://www.nvidia.com/object/producttesla_c1060us.html

Also, does anyone know how many NVIDIA GeForce GTX 285 or Quadro FX 4800 cards one can fit into a Mac Pro or if there are any versions that provide 4 GB or memory?

Mac OS X (10.4.3)

Posted on Aug 26, 2009 11:40 AM

Reply

Answer 1

The hatter

Level 9

61,031 points

Aug 26, 2009 2:00 PM in response to forman

Build a box around your application. Tri-SLI plus PhysX card seems to be it right now.

Why build one? you want 6-8 aux power connectors, and would require adding PSU at least. You have two 16x slots but then you don't have any 8x slots, only 4x.

Right now, without modification, only one 'high end' (285; FX Quadro) is supported. There is also the 300W for all PCI slots, but does that include auxillary power in that total?

There are some 2GB GTX 285s as well as 295s - PC only.

Reply

Answer 2

Aug 26, 2009 4:21 PM in response to The hatter

There is also the 300W for all PCI slots, but does that include auxillary power in that total?

Yes,
<http://developer.apple.com/documentation/Hardware/Conceptual/HWtech PCI/Articles/pciimplementation.html>
"Max total PCI Express power (slot power and aux connector power): 300 W"

Reply

Answer 3

forman Author

Level 1

4 points

Aug 27, 2009 11:29 AM in response to The hatter

So it sounds like there are power and bus limitations to adding multiple computing processors to a Mac Pro, which necessitates the use of a custom enclosure. Further, because of hardware and driver limitations, the GTX 285 is will likely be the fastest available computing processor for a while at least.

With that said, is it possible to build an enclosure with three GTX 285 cards that can be interfaced to a Mac Pro? I'm not familiar with hardware that allows one to bridge to an additional bus.

Also, what about an Xserve? I see that they can have up to two 700 W power supplies, however they only have two open x16 PCI Express 2.0 expansion slots, one of the being half length. Again, one would need to expand the number of slots.

Reply

Answer 4

Aug 27, 2009 11:37 AM in response to forman

There are systems to add external PCI-e cards but the data transfer rate may be limited. If the application does a lot of computing compared to the amount of data transferred, that might not be a problem.
<http://www.magma.com/products/pciexpress/index.html>

Reply

Answer 5

The hatter

Level 9

61,031 points

Aug 27, 2009 11:45 AM in response to forman

There use to be and likely still are PCI Express expansion boxes, but I was referring more to running a Windows 7 only rig you can build yourself. Harder to do it perfect if you want dual cpu / 8-core so you might want HP or Dell workstation.

Core i9 is due out 4th quarter from Intel, 1000-series chips and 5600-series Xeon with 6-cores and possibly 8 on single LGA-1366 socket (same as 5500 and current Core i7).

Intel and EVGA and others have X58 boards you can use to build your system around, I've got two, an Intel and Gigabyte currently. Support for 2-3 16x slots and 8x along with 1x and older PCI style slot. Plus, 6 DDR3 DIMM slots.

Reply

Answer 6

forman Author

Level 1

4 points

Sep 10, 2009 1:31 PM in response to The hatter

I would like to avoid the Windows/Dell ghetto, if at all possible. As a Unix and Mac OS X user who has spent much of his career keeping a second computer next to his primary workstation just for simulations, I am no longer interested in continuing that inconvenient and costly paradigm. Indeed, I immediately decided to purchase an XFDTD license simply for its Mac OS X support.

With that said, I understand that there are GNU/Linux solutions that support the Tesla C1060 Computing Processor, which if a solution does not exist for Apple hardware, would be the next best thing. XFDTD runs in GNU/Linux as well.

On a related note, I was told that NVIDIA engineers feel that the biggest demand for their Computing Processors is in Linux. If there are NVIDIA employees who might read this, I would like to state that Mac OS X is an extremely competent scientific computing platform and with the addition of OpenCL to the operating system both companies could benefit immensely from compatible Computing Processors. Indeed, if NVIDIA's hardware worked with EFI, I'd have ordered several already.

In October I plan to build a system to add external cards to a Mac Pro using the GTX 285 for now. Perhaps if I can build a proof-of-concept system, it will generate more interest in brining the C1060 Computing Processor to Apple hardware.

Message was edited by: forman

Reply

Answer 7

gebi

Level 1

0 points

Sep 23, 2009 3:17 AM in response to forman

Hi,
unfortunatly there is an EFI related problem with the current versions of the tesla boards.
Some folks say that they could activate a tesla board within a macpro 2009 running centos (redhat clone). The boards need both extra power »cords« on the motherboards.

Two options: use a Quadro FX 480 for mac which has at least 1.5GB of RAM. You can add up to 4 in a appropriate magma box with sufficient power from the box or use 285er instead if you like.

There is also hope but not officially. I guess there will be newer versions available with the newly announced chips from Nvidia. There is not so much to change from nvidia's side. It's simply firmware and a lowlevel driver problem.

Reply

Answer 8

The hatter

Level 9

61,031 points

Sep 23, 2009 5:35 AM in response to forman

I just saw a custom build "concept" tri-HAF932 case (looks like a solid server) with graphics cards in one case, hard drives in #2, and guts and water cooling block in main section, all bound into one case.

Can't find what I saw but this gives 2/3 the idea 🙂
http://www.evga.com/forums/tm.asp?m=100777482

eVGA 4-Way Classified+ Tesela C1060
http://www.evga.com/forums/tm.asp?m=100927093

OpenCL is getting the buzz and support, and my guess is CUDA/GPGPU has given Nvidia engineers a heads up. As for Apple... there are limits left and right compared to custom built and unlocking the cpu and gpu(s).

Nvidia's Quadro CX was designed strictly to support Adobe Suite on Windows showed a lot of promise (haven't heard anything since it was demoed).

Reply

Answer 9

The hatter

Level 9

61,031 points

Sep 23, 2009 6:15 AM in response to The hatter

Lucid promised us GPU scaling across multiple GPU generations with near-linear performance gains without restrictions of SLI or CrossFire.

*HYDRA technology.* In this article there is a quick look at the MSI 'Big Bang' motherboard that sports the P55 chipset and HYDRA chip and also shows some demos of AMD HD 4890 and NVIDIA GTX 260 graphics cards working together for game rendering.

Truly platform-independent GPU scaling is nearly here and the flexibility it will offer gamers could be impressive."

Slashdot

Reply